Activity

1. Carreau commented on Mar 14, 2020

   

   Carreau

   on Mar 14, 2020

   Member

   More actions

   I would try to update your matplotlib/ipykernel/ipython, it should be automatic since a couple of releases ago:

   
2. tdpetrou commented on Mar 14, 2020

   

   tdpetrou

   on Mar 14, 2020

   Author

   More actions

   Importing matplotlib.pyplot triggers the import ipykernel.pylabl.backend_inline. This runs the function _enable_matplotlib_integration, which imports from IPython.core.pylabtools and calls activate_matplotlib and configure_inline_support. Try running just the second cell without importing pyplot.
3. tdpetrou commented on Mar 14, 2020

   

   tdpetrou

   on Mar 14, 2020

   Author

   More actions

   Here's a summary of what happens

   What happens when jupyter notebook is launched from command line

   • When starting a new notebook, ipykernel.kernelapp.IPKernelApp is created and the initialize method calls init_gui_pylab which sets the backend to inline

   What happens when import matplotlib.pyplot is run

   1. switch_backend is called
   2. importlib.import_module(backend_name) is run
   3. ipykernel.pylab.backend_inline is imported which runs _enable_matplotlib_integration
   4. which imports from IPython.core.pylabtools and calls activate_matplotlib and configure_inline_support
   5. activate_matplotlib callsswitch_backend again, but running importlib.import_module(backend_name) doesn't trigger calling of module since its already imported.
   6. configure_inline_support registers flush_figures from ipykernel.pylab.backend_inline for post_execute event
   7. select_figure_formats is run once to register function hook for each type of image
   8. Example: png_formatter = shell.display_formatter.formatters['image/png']
   9. png_formatter.for_type(Figure, lambda fig: print_figure(fig, 'png', **kwargs))
   10. print_figure function is finally called which calls the fig.canvas.print_figure function to place data in a BytesIO stream. It is this data that is returned.
   11. In pyplot, install_repl_displayhook is called and registers event post_execute to draw_all figures - draw using the canvas

   What happens when %matplotlib inline is run

   1. Method matplotlib is run from IPython/core/magics/pylab.py
   2. Calls shell.enable_matplotlib(gui) from IPython/core/interactiveshell.py
   3. This runs from IPython.core import pylabtools as pt and calls
   4. pt.find_gui_and_backend - imports matplotlib
   5. pt.activate_matplotlib(backend) - backend set, imports pyplot
   6. importing pyplot calls ipython.core.pylabtools.activate_matplotlib again
   7. pt.configure_inline_support(self, backend) does same as above
4. tacaswell commented on Mar 15, 2020

   

   tacaswell

   on Mar 15, 2020

   Contributor

   More actions

   You do not want to import Matplotlib on notebook start up because if the user does not need it we should not waste the import time.

   We also need to be sure that these changes do not break either nbagg or ipympl.

   Steps 7-9 are the key steps for making the implicit display work.

   Probably the right way to fix this is to add a __repr_html__ to Figure in Matplotlib that then defers to a method on the canvas object. We would then put the correct methods on the canvas classes from inline (the static backend), nbagg (which we will hopefully be able to retire in the next year or two), and ipympl (the ipywidegts based notebook backend).

   Reacted by Aurélien Geron, texnokrates and felipe
5. tdpetrou commented on Mar 15, 2020

   

   tdpetrou

   on Mar 15, 2020

   Author

   More actions

   @tacaswell Agreed, _repr_html_ would be a nice way to fix this to avoid the extra imports at the start and not affect the implicit plotting.
6. timics commented on Mar 29, 2021

   

   timics

   on Mar 29, 2021

   More actions

   Came here from the 'matplotlib rcParams are not picked up from if "%matplolib inline" is in the same cell' bug, curious if anyone is still looking into either fixing that issue or getting rid of the need for "%matplotlib inline"
7. AstroMike commented on Apr 14, 2021

   

   AstroMike

   on Apr 14, 2021

   More actions

   @timics In recent versions of jupyter notebook I have been omitting %matplotlib inline entirely and it still works, so I am assuming ithas been updated to remove the need for this call.
8. westurner commented on Apr 15, 2021

   

   westurner

   on Apr 15, 2021

   Contributor

   More actions

   • DOC: what do the docs for %matplotlib inline clarify?
9. jklymak commented on Apr 3, 2025

   

   jklymak

   on Apr 3, 2025

   More actions

   Probably the right way to fix this is to add a __repr_html__ to Figure in Matplotlib that then defers to a method on the canvas object.

   Matplotlib has figure.Figure._repr_html_.

   https://github.com/matplotlib/matplotlib/blob/a364dc5295e332b2d7e038f18104d0a94aad7ef2/lib/matplotlib/figure.py#L2743-L2748

   does it not work here?